Compressed sensing reconstruction of dynamic contrast enhanced MRI using GPU-accelerated convolutional sparse coding

In this paper, we propose a data-driven image reconstruction algorithm that specifically aims to reconstruct undersampled dynamic contrast enhanced (DCE) MRI data. The proposed method is based on the convolutional sparse coding algorithm, which leverages the Fourier convolution theorem to accelerate the process of learning a collections of filters and iteratively refines the reconstruction result using the sparse codes found during the reconstruction process. We introduce a novel energy formation based on the learning over time-varing DCE-MRI images, and propose an extension of Alternating Direction Method of Multiplier (ADMM) method to solve the constrained optimization problem efficiently using the GPU. We assess the performance of the proposed method by comparing with the state-of-the-art dictionary-based compressed sensing (CS) MRI method.